To Our Clients and Friends,
The last edition of our Corporate Governance Newsletter focused on draft amendments to the Part 500 Cybersecurity Rules proposed by the New York Department of Financial Services (the “NYDFS”). The amendments include significant changes relating to governance, technology, risk assessments, notifications, and penalties.
This month, we are going to step back and think about the topic of “noise” in corporate decision-making, and the role that governance processes can play in mitigating the impact of noise.
The Impact of “Noise” in the Governance Process
In 2021, researchers and authors Daniel Kahneman, Olivier Sibony and Cass R. Sunstein published an exhaustive study of “Noise,” describing an aspect of the decision making process that had theretofore not received much attention. “Noise” in this context is defined in distinction from “bias.” Bias is a systemic leaning in a certain direction, whether explicit or implicit (so, a set of arrows all miss the bullseye but land in the same quadrant of a target). Noise is the introduction of random results into decision making (so, a set of arrows all miss the bullseye but they are dispersed across different quadrants of the target). Both bias and noise have adverse effects on decision-making, but noise is both less widely acknowledged and also harder to perceive.
As a primary example of the impact of noise on corporate decisions, the book uses a scenario drawn from the insurance industry. Specifically, the authors describe a situation in which insurance company executives were seeking to diminsh the amount of variability they were seeing across underwriting decisions. Though executives acknowledged that there was a degree of noise in their processes (the assignment of underwriters and claims managers was based on who was available rather than other, less “lottery-like” criteria), they differed in evaluating how significant the noise was. So they undertook a “noise audit” in which five different hypothetical scenarios were presented for evaluation. While this test group estimated that the impact of “noise” would create roughly 10% variability in their results (a 10% swing in pricing a risk or the amount-of a claims settlement), the actual measured impact was radically greater: 55% for underwriting decisions and 43% for claims decisions. Much of the book is devoted to considering the causes and effects of this type of random noise, and then developing systems that can mitigate its impact.
These observations got us thinking about the broader effect of “noise” on corporate decision-making and in particular on the role of governance processes in mitigating the adverse impact of noise. If a system can be built to help create greater consistency of results, without losing the benefits of diverse perspectives, intuitions and experiences, that seems like a good result.
Some specific, practical points emerge from Noise. In particular, the authors describe a decision-making process that is intended to mitigate the effects of noise. That process consists of six main steps, starting with developing a list of “mediating assessments”, meaning those determinations that will be key to deciding whether or not to pursue a project such as a material acquisition. Once the decision-making process is rooted in an agreed list of assessments, the authors suggest that much of the noise in a process can be reduced. The other aspects of this process include obtaining an outside view when possible; keeping the assessments as independent of each other as possible; reviewing each assessment separately; ensuring that participants make their judgments individually; and in the final decision process, when the assessments are all put on the table for evaluation and decision-making, allowing for the re-introduction of a degree of judgment and intuition.
While this type of process is actually not very different from that we see already being de facto implemented in complex decision-making across many of our clients, we believe the conscious effort to identify and distill the noise out of decision-making is a laudable one. We believe it is worth considering whether corporate governance processes are well-designed to limit the impact of noise by anchoring decision-making in this type of assessment-based model.
Of course the pursuit of greater consistency of decision-making can also mask implicit bias and result in a loss of the benefits of diversity in the decision-making process. We see some of this debate echoed in the regulatory questions associated with the increasing use of artificial intelligence in the insurance industry: less noise and thus greater consistency of decisions can be aided by stronger algorithms slicing data more and more finely. On the other hand, those improved results may well have unintended adverse policy consequences that the regulatory community is positioned to assess and counteract.
The concept of bias in decision-making has been discussed at length in the literature around corporate governance. The concept of noise (random inputs significantly impacting results) is less well-known. Mitigating the impact of noise on corporate decision-making seems intuitively like a good and important goal, resulting in better and more consistent decisions. Governance processes that are rooted in a “mediating assessment”-based protocol may well help limit the adverse impact that the authors of Noise observed when evaluating an insurance company’s underwriting decisions.
However improved decision-making that mitigates the effect of noise may also run up against the policy objectives of regulators - not that regulators embrace noisy decisions, but they do seek to ensure fairness and create equitable outcomes across classes of policyholders.
Perhaps the way to balance these competing objectives is, as many of our clients have already observed, to include the regulatory response to a proposed decision among the relevant “mediating assessments” - a clear-eyed view of the likely regulatory reaction to an acquisition or other material corporate decision can and should be a part of a decision making process that is rooted in an objective and relatively noise free process.